From Separation to Participation: Predicting Reserve Service Outcomes Using Machine Learning in the Australian Defence Force

Abstract

The Australian Government has directed the Australian Defence Force (ADF) to increase its Reserve workforce by 1,000 additional members by 2030; however, current Reserve recruitment and full-time-to-Reserve transition rates remain insufficient to meet this target. This thesis uses individual-level demographic, geographic, and economic data to train and evaluate machine learning models that predict Service Category (SERCAT) choice, subsequent Reserve participation and service intensity among separating full-time ADF members from FY2016–17 to FY2024–25. The models demonstrate strong performance in classifying SERCAT choice but substantially weaker performance for downstream participation and intensity outcomes. Institutional and geographic characteristics are most influential at the point of separation, whereas personal and life-stage characteristics are more strongly associated with post-separation Reserve engagement. These findings demonstrate how machine learning can inform targeted Reserve outreach strategies to support directed workforce growth, while also revealing structural and individual constraints that limit realized participation after separation.